The mean is probably the average cloud cover in percentages: 41% cloud coverage on average over time. The categories you mentioned: 'clear', 'scattered', 'broken' and 'overcast' are defined in different ways sometimes. Basically there is a human observing the sky analyzing the coverage in oktas or tenths (e.g. 6/8), or more likely there is a sounding device looking through the sky and getting a value in percentages of the cloud cover (e.g. 75.0%). To get one of these categories from the device data, the percentages are turned into oktas or tenths. Some data is lost in this transformation and thus from the categorized data it is impossible to get calculate the actual mean.
Some use more the oktas (World Meteorological Organization) and others (I would guess often USA) tenths in the categorization. Also after quick googling the categorization bins seem to be differ from source to source at least for the ones who use tenths. The clear and overcast bins are for all the same less than 1/8 or 1/10 and more than 7/8 or 9/10 respectively. Obscured also always means basically that the value is undetermined: perhaps there is fog and the device doesn't get the responses it would've wanted. Some have also a category 'few', but since it was not present in the data they have used a categorization that doesn't include it.
Based on all above I would guess the categorization is the same as in this webpage. Categories are clear: less than 1/10, scattered: 1/10 to 5/10, broken: 6/10 to 9/10 and overcast: over 9/10. As the tabulated values are historical they give a climatological average of what are the possibilities of different cloud cover scenarios. I would suspect that clear and scattered scenarios are good enough for watching an eclipse so e.g. in Salem you would have a 32.2 + 19 = 51.2 percent chance to have a decent or better viewing conditions.